Merge and split discrete cosine block transform method

Abstract

Discrete Cosine Transforms in a radix-2 block transform method enables true split and merge transform processing of equal sized data halves in the transform domain for enabling recursive merges and splits in transform domain without data degradation. Input data in the time domain or spatial domain during either the split and merge radix-2 forward transform processing is firstly transform processed followed by combinational processing. In the split transform process, whole transformed data is split using combinational processing into first and second transformed data halves in the transform domain as a true split. In the merge transform process, first and second transform halves are merged using combinational processing into a merged transformed data whole in the transform domain. In either case, time or spatial domain input data can be transformed into the transform domain in the form of split halves or merged whole. The split halves can be merged by the merge process combinational processing and the merged whole can be split by the split process combinational processing. The split and merge process combinational processing enable recursive splits and merges in the transform domain.

Description

REFERENCE TO RELATED APPLICATIONS[0001]The present application is one of six related copending applications including a base application entitled Merge and Split Generalized Block Transform Method, Ser. No. 10 / 175,388 and five specific applications entitled Merge and Split Fast Hartley Block Transform Method, Ser. No. 10 / 174,894, Merge and Split Fast Fourier Block Transform Method, Ser. No. 10 / 177,197, Merge and Split Discrete Cosine Block Transform Method, Ser. No. 10 / 175,594, Merge and Split Discrete Sine Block Transform Method, Ser. No. 10 / 174,925, Merge and Split Karhunen-Loeve Block Transform Method, Ser. No. 10 / 175,397, all filed Jun. 19, 2002, by the same inventor.FIELD OF THE INVENTION[0002]The invention relates to the field of separate and combine block transforms for communicating data. More particularly, the present invention relates to a new class of true merge and split forward transforms for lossless recursive merging and splitting of transformed data in the transform ...

AI Technical Summary

Benefits of technology

[0013]An object of the invention is to recursively split data into transform halves without degrading data quality.

[0014]Another object of the invention is to recursively merge data halves into a transform whole without degrading data quality.

[0016]Still another object of the invention is to provide a generalized transform method, that can be applied to Fast Fourier Transforms (FFTs), Fast Hartley Transforms (FHTs), Discrete Cosine Transforms (DCTs), Discrete Sine Transforms (DSTs), and the Discrete Karhunen-Loeve Transforms (KLTs), that can split data into transform halves without degrading data quality, that can merge data halves into a transform whole without degrading data quality, that can merge the transform halves into the transform whole without data degradation and without inversely transforming the transform halves, and that can split the transform whole into the transform halves without data degradation and without inversely transforming the transform whole, and that can recursively back-and-forth merge and split transform data in the transform domain.

[0017]The present invention is directed, in a generalized approach, to true direct split and merge transform processing of equal sized data halves in the transform domain. The invention, in the general aspect, is directed to a new class block transforms that enables true successive and back-and-forth merge and split forward transformations without data degradation. Input data in the time domain or spatial domain during either split or merge radix-2 forward transform processing is firstly transform processed followed by combinational processing. In the split transform process, input data is split into first and second transformed data halves in the transform domain as a true split. In the merge transform process, first and second time domain data halves are merged into a merged transformed output. In either case, time or spatial domain input data can be transformed into the transform domain in the form of split halves or merged wholes. The split halves and merged wholes can then be successively recursively further split or merged in the transform domain without inverse transformation and without resulting data degradation. That is, the merge and split transformation processes can be used to merge and split transform data without having to inversely transform transformed domain data back into the time or spatial domain.

[0018]Referring to FIG. 1A, the present invention in the general form enables true split and merge operation in the transform domain using unique processing architectural characteristics. A first unique characteristic is that both merge and split processes firstly front-end transform process the input data into transform data and then secondly back-end combinationally CLA and CLB process the transform data into merged transformed whole data or split transformed half data, respectively. The second unique characteristic is that the split process uses a whole G transform and the merge processes uses radix-2 G / 2 transforms of the whole transform as a true radix-2 transform process. A third unique characteristic, flowing the first and second characteristics, is that the back-end combinational processing can be cross-fed between the merge and split combinational processing for true recursive splits and merges in the transform domain. Particularly, the split transformed data halves output from the split combinational logic CLA processing can be cross-fed into the merge combinational logic CLB processing for merging the two transformed halves into the merged transformed whole data, and the merged transformed whole data can be cross-fed into the split combinational logic CLAprocessing for splitting the merged transform whole data into the split transformed halves data. The cross-fed radix-2 recursive true merge and split transformation capability enable manipulation of data in the transform domain without data degradation.

[0019]Referring to FIGS. 1B and 1C, the general block transform processed performs true split and merge block transform processes. For example, transformed output data of the split block process shown in FIG. 1B, is split into first and second transform halves G(Xp) and G(Xr) that can then be an input into the combinational processing of the merge block process for merging the first and second transform halves G(Xp) and G(Xr) back into a merged transformed output as shown in FIG. 1C. For another example, the merged transformed output data of the merge block process of FIG. 1C is a merged transformed output of even and odd halves Ze and Zo that can then be an input Zo and Ze into the combinational processing of the split block process for splitting merged transformed domain output data Zo and Ze back into first and second transform halves G(Xp) and G(Xr) using the combinational processing of the split block process of FIG. 1B. The split transform data can be recursively split, and merged transformed data can be recursively merged. Merged and split data can be back-and-forth recursively merged or split, all in the transform domain, without data degradation, and without inversely transforming the forward transformed data. Two transformed data halves can be merged into the merged transformed data that can be split back into the two transformed half data without data degradation occurring and without inversely transforming the transformed data halves. Likewise, the merged transformed whole data can be split into the two transformed half data and then merged back into merged transformed data without data degradation and without inversely transforming the transformed data whole. The radix-2 split transformed data can be successively recursively split again and again with each split being a radix-2 split, to the radix-Nth split where N is number of the power of two, and each split is generated using smaller ½ size combinational processing. For example, the radix-2 split transform data can be fed to a CLA / 2 split combinational process splitting the radix-2 split transformed data into half size radix-4 split transformed data. The half size radix-4 split transformed data can be further split again, or merged, for example, by a half size CLB / 2 combinational process back into the original radix-2 split transformed data. As may be apparent, CLA is a double size split combinational process being twice the number of points of the CLA / 2 combinational process, and, CLB is a double size merge combinational process being twice the number of points of the CLB / 2 half size combinational process. Recursive splits are recursive successive splits and recursive merges are recursive successive merges. Recursive splits and merges are recursive back-and-forth splits and merges. The transformed data can be recursively, either successively or back-and-forth, split and merged by half size and double size merge and split combinational processes. The primary benefit of the invention is the ability to perform recursive back-and-forth splits and merges in the transform domain without data degradation nor having to inversely transform the data.

Problems solved by technology

The traditional methods that perform separation and combination on the transformed data blocks require that the individual blocks be inversely transformed back to the spatial domain because these existing fast transforms do not have the capability of performing true merge and split strictly in the transform domain.

The quantization process produces degradation loss of image quality.

Manipulation in the transform domain results in further degradation of image quality.

However, each time that data is transformed, image quality is degraded.

Without true merge and split that would enable convenient repetitive data manipulation in the transform domain, data is disadvantageously transformed back into the time or spatial domain.

After forward transform domain processing, the transform domain data can be then be inversely transformed back into the original time domain or spatial data with a degradation in data quality when transforming the data due to quantization of the input data set.

The inverse transformations and retransformation can result in additional degradation of data quality.

Rather, the prior art teaches a potpourri of disjointed separate and combine radix-2 transform processes that are not true merge and split transforms, but rather separate and combine transformation that disadvantageously require additional inverse transformations and data degradation when separating and combining transform data.

The prior art transforms do not provide true merge and split transform processing because of the required inverse transformations to perfect a combination after a separation, or a separation after a combination of the transformed data.

For example, existing fast radix-2 Discrete Cosine Transform algorithms are unable to merge or split data blocks strictly in the DCT domain.

Consequently, data is degraded while inversely transforming and data can not be conveniently manipulated in the transform domain without further data degradation.

Images